CVE-2025-46153

EUVD-2025-31108
PyTorch before 3.7.0 has a bernoulli_p decompose function in decompositions.py even though it lacks full consistency with the eager CPU implementation, negatively affecting nn.Dropout1d, nn.Dropout2d, and nn.Dropout3d for fallback_random=True.
ProviderTypeBase ScoreAtk. VectorAtk. ComplexityPriv. RequiredVector
NISTPrimary
5.3 MEDIUM
NETWORK
LOW
NONE
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:N/A:N
Base Score
CVSS 3.x
EPSS Score
Percentile: 32.44%
Affected Products (NVD)
VendorProductVersion
linuxfoundationpytorch
2.6.0 ≤
𝑥
< 2.7.0
𝑥
= Vulnerable software versions
Debian logo
Debian Releases
Debian Product
Codename
pytorch
bookworm
no-dsa
bullseye
postponed
bullseye (security)
vulnerable
forky
2.12.1+dfsg-1
fixed
sid
2.12.1+dfsg-1
fixed
trixie
no-dsa
Ubuntu logo
Ubuntu Releases
Ubuntu Product
Codename
pytorch
jammy
needs-triage
noble
dne
plucky
ignored
questing
ignored
resolute
needs-triage
Common Weakness Enumeration
References
https://gist.github.com/shaoyuyoung/4bcefba4004f8271e64b5185c95a248a
https://gist.github.com/shaoyuyoung/e636f2e7a306105b7e96809e2b85c28a
https://github.com/pytorch/pytorch/compare/v2.6.0...v2.7.0
https://github.com/pytorch/pytorch/issues/142853
https://github.com/pytorch/pytorch/pull/143460